32nd Annual International Conference of the IEEE EMBS

Buenos Aires, Argentina, August 31 - September 4, 2010

Extension Assist Control for Individuals with Cervical Cord Injury

Using Motion Assist Robot for Upper Limb
Tatsuya Watanabe and Kenichi Yano

Abstract Many people of all ages have sustained cervical
cord injury in traffic accidents or sport accidents, and conse-
quently suffered physical impairment. Among these individuals
dysfunction of the upper limbs is a concern, although recovery
from dysfunction is possible through rehabilitation. In this
study, we developed an assistive robot for upper limb movement
which has high effectiveness in rehabilitation. To achieve this,
we devised an algorithm of a dynamic filter that decreases the
noise and delay of the device for controlling the motion assist
robot stably. This filter changes the cutoff frequency depending
on the amount of input change. The robot assists movement
only during elbow extension. The effectiveness of the proposed
method is demonstrated through experiments.
of restraint can result, owing to the dynamics of the device
and the delay of motion.
In this study, we propose a novel dynamic filter that fo-
cuses on reduction of input delay. By improving the transient
response performance, we aim to achieve high usability. We
also aim to allow control of assistance only during extension
motion. Furthermore, we make gain adjustments according
to the angular velocity of the upper limb, such that the user
does not experience delay during movement.
User

I. INTRODUCTION Non-Assist

Recently, the annual incidence of spinal cord injury (SCI) Pressure Sensor
has been reported to be approximately 12,000 new cases
each year in the United States[1]. In Japan, meanwhile, the
number of people with SCI has been estimated to be around Digital Filter Assist
100,000, and individuals with cervical cord injury (CCI) are
thought to account for about 60% of SCI cases. Two out of
Actuator
three cases of SCI are caused by traffic accidents or falls.
Motion-Assist Robot
If the cervical cord, which is a component of the central
nervous system, is damaged by trauma to the cervical spine Computer AD/DA
near the brain, which is located above the cervical vertebrae,
Fig. 1. Schematic diagram of motion assist robot for upper limb
CCI can occur. The cervical spine contains eight cervical
nerves: C1 to C8. For example, if C3 is damaged, spinal
cord function at C4 and below may be lost, leading to II. CERVICAL CORD INJURY
paralysis. If injury to C6 occurs, the ability to extend the
As reported in a number of studies, the stiffness of the
elbow may be lost, owing to paralysis of the upper limb, in
human body changes according to the velocity of motion
addition to the lower limb. Consequently, activities of daily
[5][6]. In particular, finely tuned adjustment of ones body at
living (ADL) become severely limited for such individuals.
low velocity is made possible by increasing stiffness through
Furthermore, individuals with partial paralysis experience
the action of antagonistic muscle. For instance, if the triceps
difficulty in continuing rehabilitation at home, and are at
brachii applies force in the direction of extension, the speed
high risk of functional deterioration due to limited range
of the motion can be finely tuned by biceps brachii, which is
of motion and muscular atrophy. Therefore, to facilitated
the corresponding antagonistic muscle. However, the targets
active rehabilitation at home and thereby prevent loss of the
of this study, namely, individuals with C5 level CCI, have
functional abilities retained by disabled individuals, a motion
paralyzed triceps brachii; thus, these individuals can extend
assist robot for the upper limb is necessary.
their elbow only by relaxing their biceps and utilizing the
To address these issues, many wearable motion assist force of gravity. Control of motion in this manner does not
robots have been studied to date. For devices controlled allow for smooth extension of the elbow. Furthermore, the
by pressure sensors that detect contact between the device triceps brachii is not used as the antagonistic muscle during
and the wearers body[2][3][4], it is necessary to maintain elbow flexion.
close contact between the wearer and the device, to ensure In this study, the subject was a 30-year-old male with C5
accurate detection of operational commands and to stabilize level CCI. The subjects motion during elbow extension was
displacement of the contact part. However, if part of the investigated by using motion capture. We asked the subject
motion assist robot is in close contact with the user, a sense to perform elbow extension as slowly as possible (Fig. 2).
This work was supported by City Area Program (Development stage) : For individuals without injury, the velocity of the extension
Southern Gifu Area. movement is constant when the elbow is extended slowly.
978-1-4244-4124-2/10/$25.00 2010 IEEE 1312
 Angle [deg] 0

-20

-40

-60

0 1 2 3 4 5 6
Time [s]

30
Angular Velocity [deg/s]

20

Fig. 3. Motion assist robot for upper limb

10

Input Force [V]

Raw Data fin
0 5
Low-Pass Filter(10Hz) fLPF
0
-10
0 1 2 3 4 5 6 -5
Time [s]
0 2 4 6 8
Time [s]
Fig. 2. Angle and angular velocity during attempt by individual with C5
level CCI to perform elbow extension as slowly as possible 20
Angle [deg]

However, the results presented in Fig. 2 show that the subject -20
could not adjust the angle of extension smoothly; moreover, 0 2 4 6 8
the angular velocity was sometimes 0, signifying that the Time [s]
Potentiometer [deg]

arm stopped at certain points. Clearly, the subject could not

finely tune the upper limb motion using the triceps and could 0
not apply force during elbow extension. -0.02
-0.04
III. WEARABLE MOTION ASSIST ROBOT FOR -0.06
UPPER LIMB 0 2 4 6 8
Time [s]
Figure 3 shows the wearable motion assist robot for the 2
Output [V]

upper limb developed in this study. The device is wearable,

and a servo motor is built into the cubital region. The device 0
is controlled by input measured by pressure sensors. This
robot is divided into two parts: the inner part fitted to the -2
0 2 4 6 8
wearer and the outer part driven by the actuator. The inner Time [s]
part is able to operate freely. The rotation axis of both the
users elbow and the inner part corresponds to the rotation Fig. 4. Input signal and motion of device
axis of the outer part on this device; therefore, sensors can
detect inputs without problems if extension or flexion is
performed while wearing the device. IV. CONTROL DESIGN
We conducted an experiment to verify the developed In this study, control is based on analog inputs; thus, noise
motion assist robot for the upper limb. The basic control from various sources might be generated before the signal
technique for this robot is admittance control based on enters the controller. It is difficult to remove all noise while
signals from pressure sensors. Figure 4 shows the results of retaining the input signals. In addition, input delay affects
the verification experiment for extension and flexion of the the stability of control because sensors are always in contact
elbow while wearing the device. Signals from the pressure with the wearer of the device. Therefore, we attempted to
sensors were run through a low-pass filter (LPF) with a 10 Hz design a dynamic filter (see Fig. 5) with effectiveness for
cutoff. As a result, the robot could be controlled well during reducing the input delay, as well as the same effectiveness
operation; however, the device also vibrated continuously as the low-pass filter.
after stopping the extension or flexion motion. Td [s] is a time constant of the filter. The cutoff frequency
1313
 0.3
Raw Data fin
f in f Dy
1 0.2 Dynamic Filter fDy
Td s + 1
0.1

Input Force [V]

f Dy
s
Hd Tlpf s + 1 0

s -0.1
0.1s + 1
-0.2
Fig. 5. Block diagram of dynamic filter

0 0.1 0.2 0.3 0.4 0.5 0.6

of the filter was set as shown in Eq. 1. We consider that noise Time [s]
at the low cutoff frequency can be removed if Td is much Fig. 7. Result of passing input signal through dynamic filter under static
conditions
larger, but the response delay might then increase because Td
is a time constant. fin [V] is the input signal from a pressure
sensor, and A [-] and B [-] are constants. In this study, we 8 Raw Data fin
set A=1.01 and B=50.0. LPF(10Hz) fLPF

Input Force [V]

6 Dynamic Filter fDy
1
= Hd ( fDy ) = log10 (| fDy | + A)B (1) 4
Td
For the evaluation of the cutoff frequency, the derivative 2
value is passed through the LPF as shown in Eq. 2, because
the value can deviate if fDy is differentiated directly. 0
 0 0.1 0.2 0.3 0.4 0.5
1 100 ( fDy 0) Time [s]
= (2)
Tl p f 10 ( fDy < 0)
10
Here, we consider the difference between using a dynamic
filter and using a LPF with a 10 Hz cutoff frequency for input
Angle [deg]

signals from pressure sensors. 0

0.3
Raw Data fin -10
0.2 Low-Pass Filter(10Hz) fLPF

-20
0.1 0 0.1 0.2 0.3 0.4 0.5
Input Force [V]

Time [s]
Fig. 8. Comparing of rise time of input passing filters
0

-0.1
Figure 8 shows the results when each input signal is passed
-0.2 through either a dynamic filter or LPF. The subject who
wore the device extended his elbow momentarily to impart
an impact force that was as large as possible. Comparing
0 0.1 0.2 0.3 0.4 0.5 0.6 the signal passed through the dynamic filter and the raw
Time [s]
Fig. 6. Result of passing input signal through low-pass filter under static
data, the delay of the input passed through the dynamic
conditions filter is less than about 15 ms, while the delay for the signal
passed through the LPF is about 100 ms. Even considering
Figure 6 and 7 show the results under static conditions the peak value, the difference between the raw data and
where an able-bodied individual wore the device and main- the input passed through dynamic filter was approximately
tained a constant arm position. Under these static conditions, 0.20 V, while the input signal passed through the LPF was
it is desirable that the input signal maintain a constant value approximately 2.0 V. Thus, the dynamic filter causes little
of zero. Both the LPF and the dynamic filter prevented delay. In comparison with an LPF, the proposed method
approximately 0.1 [V] of the noise and allowed a stable can reduce input delay and allows for fine control with low
value to be maintained. vibration.
1314
 To obtain stable operations, the impedance parameter was 10
Actual Input fDy
set depending on the angular velocity of the upper limb. System Input Ga(fDy)

Input Force [V]

5
Equation 3 shows how the damper coefficient is adjusted. In Flexion
this study, we set the initial viscosity as Ds =0.20 Nm/(deg/s) 0
and the viscosity decrease rate as AD [-]=5.0.
1 -5
DI = Ds (3) Extension
|h |
AD +1 -10
0 1 2 3 4
Time [s]
|h | [deg/s] is the angular velocity of the wearers arm, a
potentiometer is built into the device to measure the angle
20
difference of the inner part and the outer part. If motion is
stopped, the viscosity is increased to reduce the vibration of 10

the device. If motion is initiated, viscosity is decreased to
provide a feeling of responsiveness and lightness.
In general for an individual with C5 level CCI, the force
is set depending on whether force is applied by the biceps in
the flexion direction or whether this force is reduced, because
the triceps is paralyzed. When the device is operated, the
assisting force is applied only during extension. The amount
of force is set according to Eq. 4.
max(h )
fDy + (arctan( 40 ) As ) (h 5)
fa = fDy + (arctan( h ) Bs ) (h 5) (4) that the device can be controlled by small user input while
100
fDy (5 h 5)
The assisting force that causes the maximum value of h
can be freely adjusted each time. If motion is slow, the as-
sisting force would be larger when max(h ) is small in terms
of stiffness. However, the assisting force in the direction of
extension becomes larger when the higher velocity of motion
is desired because exerting suitable force from the start can
place a burden on the user. In order to ensure safe operation
of the device, As [-] and Bs [-] are set as constants that are
upper limits on the amount of assistance. In this paper, we set
As =5, Bs =3. An individual with CCI needs assistance when
performing elbow extension, we take arctan as a function to
make the curve describing the amount of assistance steep.
V. EXPERIMENTAL RESULTS
We conducted an experiment to verify the operation of
the control method presented in this paper. The subject was
an able-bodied male in his 20s. Figure 9 shows the result
of when the subject performed extension and flexion of the
elbow more slowly than the motion in daily life.
In Fig. 9, the value of the input of pressure sensors passed
through the filter is shown by the dashed line. The input to
the system for calculating the amount of assistance is shown
by the solid line.
Large variation in the angle of the device was not seen
in the directions of extension and flexion. Motion is largely
unassisted during flexion. When the subject applied operating
input in the direction of extension, fDy is added for motion
in the direction of flexsion after delay of 0.30 ms before
applying the assisting force. The applied force is in the
direction opposite to the movement, and the device was
found to push the arm down. However, this result shows
1315
Angle [deg]
0
-10
-20
0 1 2 3 4
Time [s]
Fig. 9. Operating results for assistive movement
providing assistance. In addition, the extension operation is
not hindered.
The user can stop the extension movement by applying
force in the flexion direction. This system is designed to be
appropriate to the needs of individuals with CCI.
VI. CONCLUSION
In this study, we were able to reduce the time delay of
operation by applying a dynamic filter to the input from
pressure sensors. The proposed system can be controlled in
a manner that is suitable for individuals with C5 level CCI.
R EFERENCES
[1] A. the National Spinal Cord Injury Statistical Center, Birmingham.
(2009) Facts and figures at a glance. FactsApr09.pdf. [Online]. Avail-
able: https://www.nscisc.uab.edu/public content/facts figures 2009.aspx
[2] K. Yano, J. Hashimura, T. Aoki, and Y. Nishimoto, Flexion-extension
motion assistance using an upper limb motion-assist robot based on
trajectory estimation of reaching movement, 31st Annual International
IEEE EMBS Conference, pp. 45994602, Sept. 2009.
[3] K. Kazuo and R. A. Gopura, Design of an exoskeleton for human wrist
motion assist, JSME annual meeting 2007(5), vol. 272, pp. 117120,
Sept. 1996.
[4] P.Letier, M.Avraam, S.Veillerette, M.Horodinca, M. Bartolomei,
A.Schiele, and A.Preumont, A 7-dof portable arm exoskeleton with
local joint control, 2008 IEEE/RSJ International Conference on Intel-
ligent Robots and Systems, pp. 35013506, Sept. 2008.
[5] D. Bennett, J. Hollerbach, Y. Xu, and I. Hunter, Time-varying stiffness
of human elbow joint during cyclic voluntary movement, Expeimental
Brain Research, vol. 88, pp. 433442, Apr. 1992.
[6] H. Gomi and M.Kawato, Equilibrium-point control hypothesis exam-
ined by measured arm-stiffness during multi-joint movement, Science,
vol. 272.